[Quantlib-dev] Some thoughts on MC

[Gheury E. \(DBB\) <edm...@de...>]

Hi all,

I recently ran into some troubles using the quantlib MC pricing classes.


1. stochasticprocess: drift and diffusion are currently timestep
independent. For instance, providing the timestep could lead to a more
accurate drift, as already stated in blackscholesprocess.hpp. Is there
something preventing the implementation of this idea by adding a third
parameter "dt" in the drift and diffusion prototype in
stochasticprocess.hpp ?

Should I send a code submission for this ?

2. Use of historical data for path dependent pricing in MC. I read in
the TODO page that negative values usage in the TimeGrid class should be
investigated. While I thought of using negative values in the TimeGrid
at first for historical data, I finally decided not to do it because
TimeGrid seemed to be oriented toward the MC simulation process
involving the path *generators*. Therefore, I decided to take historical
data into account in the PathPricer class by merging the historical data
and generated path into a homemade class. The passing of historical data
to the path pricer is made through derived Instrument::arguments
classes. The solution I implemented is still a a very early devlopment
stage but could be separated as follows: the use of a historical class
for historical process values (already exists: TimeSeries) and the use
of a class that merge historical data and the generated Path/MultiPath
class at the object construction.

Does this seem to be a good idea (or at least not too bad idea) ? Are
there better ways to achieve this ?

Best regards,

Edmond

=09


--------------------------------------
Dexia Bank disclaimer :
http://www.dexia.be/maildisclaimer.htm
--------------------------------------=20
  =20